自駕車於街道場景之即時防碰撞閃避路徑規劃與控制系統設計

Abstract

近年來，自駕車的控制系統不斷改良且相關技術已愈發成熟，已有越來越多的自駕車上路測試及營運，但是國外開發的系統不一定適用於國內特有交通情境上，尤其在國內街道場景中，可經常看到路邊臨時停車以及機慢車沿路緣低速行駛的案例，因此自駕車須能進行閃避的駕駛行為避免碰撞意外發生。本論文目標為發展一套可應用於自駕車閃避障礙物控制系統，在車輛行駛過程中可評估該障礙物移動模式，判定為動態或靜態障礙物，再決定以何種軌跡規劃策略來進行閃避駕駛行為。首先利用自駕車安裝之感測器判斷前方行駛路徑上是否存在障礙物，並利用障礙物動態偵測評估該障礙物動態屬性。若須閃避的障礙物為靜態障礙物，可利用分段規劃路徑以替代重複規劃相似路徑；若為動態障礙物，則會根據感測之動態資訊即時規劃新的閃避路徑，以保證車輛閃避過程中防止碰撞發生。本論文先採用CarSim與Python進行協同模擬，驗證靜態障礙物與動態障礙物情境下的閃避控制設計。最後透過實車收集人類駕駛應對障礙物的閃避軌跡與數據，並與本論文提出之軌跡進行分析與比較。實驗結果驗證了該方法有效提高計算軌跡的效率，並且所展現之閃避策略效果與人類駕駛非常接近。In recent years, the self-driving control system has been continuously improved and related technologies have become more mature. more and more self-driving cars have been tested and operated on the road. However, the system developed abroad is uncertain and suitable for the unique domestic traffic situations. Especially in domestic street scenes, temporary parking on the side of the road and slow cars driving at low speed along the curbs can often be seen. Therefore, self-driving cars must be able to evasive driving behavior to avoid collision accidents. The goal of this thesis develops a control system that can be applied to self-driving cars to avoid obstacles. During the driving process of the vehicle, the movement mode of the obstacle can be evaluated and determined as a dynamic or static obstacle, and then determine to execute which trajectory planning for avoidance driving behavior. Firstly, using the sensors installed in the self-driving car to determine whether there is an obstacle ahead of the driving path, and use the dynamic detection of obstacles to evaluate the dynamic properties of the obstacles. If the obstacle must be avoided is a static obstacle, the segmented planning path can be used to replace the repeated planning of similar paths; If it is a dynamic obstacle, a new avoidance path will be planned in real-time according to the sensed dynamic information to prevent collisions during vehicle avoidance. In this paper, CarSim and Python are used for collaborative simulation to verify the avoidance control design under static and dynamic obstacles. Finally, the trajectory and data of human drivers responding to obstacles are collected through real vehicles, and the trajectories proposed in this paper are analyzed and compared. The experimental results verify that the proposed method effectively improves the efficiency of calculating trajectories, and the evasion strategy effect shown is very close to human driving.